1 |
Hi François, |
2 |
|
3 |
> On 3 Apr 2019, at 22:51, Francois Bissey <francois.bissey@×××××××××××××.nz> wrote: |
4 |
> |
5 |
> Hi, |
6 |
> |
7 |
> As someone who was involved until very recently in installing software |
8 |
> on the New Zealand national facility I feel I should take exception to some |
9 |
> of these comments. |
10 |
> To put things in perspective |
11 |
> 1) I am using Gentoo since 2003 |
12 |
> 2) I am a regular contributor to the science team and maintain sage-on-gentoo |
13 |
> 3) I pushed for a while to have prefix working on ppc64 (the hardware was |
14 |
> at some time part of the national facility above) |
15 |
> 4) I have contributed code to spack and help fix some issues with libtool |
16 |
> in spack and occasionally suggests fix to some packages. |
17 |
> |
18 |
> Gentoo prefix is awesome but some areas are not as flexible as spack. |
19 |
> Mainly because it is designed like a gentoo distro as a single tree |
20 |
> install. Everything goes into one prefix. |
21 |
> What spack allows you to do (and that is a usual requirement): |
22 |
> allow and maintain an unhealthy forest of softwares: |
23 |
> 1) across several versions |
24 |
> 2) across various compilers |
25 |
> The whole dynamically loadable via “modules”. Each bits in its own bubble. |
26 |
> This also has limitation of course. |
27 |
> Gentoo has slots that does multiple versions of some software but it is |
28 |
> not a universal feature (nor should it be on the point of view of a distro). |
29 |
> Basically if you want to reproduce some the scenarios managed by spack you |
30 |
> need multiple prefix. |
31 |
> That’s not to say spack wouldn’t benefit from a dose of gentoo and vice versa. |
32 |
> But some Gentoo features have been voluntarily avoided :( |
33 |
> |
34 |
> Now prefix was very useful to me to offer a base userland on top of SLES 11.1 |
35 |
> (which couldn’t be updated for various reasons) that was much more recent and |
36 |
> feature-full than I would otherwise have had access too. And then I could put |
37 |
> something like spack on top if I wanted to. |
38 |
|
39 |
This scenario is exactly what I was proposing in the HSF packaging meetings, to |
40 |
have a base userland installed with portage on top of which one could manage |
41 |
the end user applications with spack if need be, or nix, or any other tool. |
42 |
|
43 |
I think stacked prefix can solve a good part of the combinatorial use case that |
44 |
spack was created for, but for now the best solution for such cases is probably |
45 |
a combination of prefix and another package manager aimed at combinatorial installs. |
46 |
But an overlay with slotted scientific packages could also be interesting to explore |
47 |
(i.e. the slot could be used to just adjust the prefix to $EPREFIX/lib/<pkg>/<version>). |
48 |
That would probably work well enough for gromacs and other similar tools. |
49 |
|
50 |
Cheers, |
51 |
-Guilherme |
52 |
|
53 |
> |
54 |
> François |
55 |
> |
56 |
>> On 4/04/2019, at 07:57, Guilherme Amadio <amadio@g.o> wrote: |
57 |
>> |
58 |
>> Hi Jon, |
59 |
>> |
60 |
>>> On 3 Apr 2019, at 12:56, Jon Woodring <woodring@××××.gov> wrote: |
61 |
>>> |
62 |
>>> Looking at the GSOC, I noticed that it’s mentioned that one of Prefix’s goals is to bring Gentoo to HPC, and actually that’s where I was trying to use Prefix. |
63 |
>>> |
64 |
>>> I don’t know if you’re familiar with Spack https://spack.io/, but I was exploring using Prefix and portage, because it has a larger community and more features. |
65 |
>> |
66 |
>> Yes, I’m advocating for using prefix for HEP (at CERN) and HPC in the HSF packaging group, |
67 |
>> but I think that they are unfortunately more interested in using spack, even though it |
68 |
>> doesn’t seem to be mature enough for what is its intended use. In any case, since you are |
69 |
>> from LANL, if your cluster has CVMFS mounted (i.e. /cvmfs/sft.cern.ch), then you can already |
70 |
>> use prefix! I have prefix installed in CVMFS, which I discussed at CHEP: |
71 |
>> https://indico.cern.ch/event/587955/contributions/2938043/ |
72 |
>> |
73 |
>> Just run /cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/x86_64/startprefix to get started. |
74 |
>> |
75 |
>> In principle, there’s nothing preventing you from using ACCEPT_KEYWORDS=‘x86_64’ in your |
76 |
>> prefix configuration. It’s just not tried by anyone yet. We all use ~x86_64 for now for |
77 |
>> prefix on Linux. On Mac OS X there’s no stable keyword, ~*-macos are the only ones. |
78 |
>> |
79 |
>> My first talk about prefix for HSF packaging group: |
80 |
>> https://indico.cern.ch/event/672745/ |
81 |
>> |
82 |
>> Other related links: |
83 |
>> https://hepsoftwarefoundation.org/workinggroups/packaging.html |
84 |
>> https://groups.google.com/forum/#!forum/hsf-packaging-wg |
85 |
>> https://indico.cern.ch/category/7975/ |
86 |
>> |
87 |
>> Cheers, |
88 |
>> -Guilherme |
89 |
> |